slopcamera

Slopcamera is a local visual studio aimed at coding agents rather than human-only GUI work. The README describes it as a tool that lets you describe a finished result to Codex, Claude or another coding agent, which then uses a Bun CLI, a TypeScript SDK and a version-matched Agent Skill to inspect sources, direct cameras, edit a composition and render output. A separate MCP server exposes a fixed diagram and image toolset. There is no Slopcamera account or hosted project database. Core idea: keep the source that owns each creative decision. Native scenes own rigs and simulations; portable scenes own supported geometry, cameras and media surfaces; diagrams own objects and labels; video projects own cuts and delivery. Rendered frames and portable assets are treated as derivatives, so a video of a character does not retain its rig. Normal media edits record cuts, timing, framing, captions and effects as project decisions, and preview and final renders use the same timeline and composition. Capabilities described in the README include: - Authoring scenes and directing cameras: place geometry, images, video, diagrams and text in a portable scene; inspect stable part IDs; make typed edits; select a named camera; render frames, contact sheets or video. A Three.js profile covers calibrated cameras, explicit animation, supported GLB geometry and a selected hardware GPU path. Spark admits saved splat worlds for local camera direction. - Native film and educational animation: the current source build can direct Blender for sets, materials, lighting, skinned characters, cloth and liquid caches; CadQuery for parametric solids and STEP; and Manim Community for mathematical animation. Seven editable starters are mentioned, including product, character, shaded street, cloth, liquid, CAD bracket and educational presenter. - Diagrams and motion graphics: editable diagrams with tldraw, SVG and PNG outputs; local raster-to-SVG vectorization with VTracer; animation of graphic layers with HTML, SVG, Motion, p5, Two, Paper Shaders or Three.js. An optional vgpu example renders programmable WebGPU passes into retained raster frames. - Media generation and direction: discover image, video, speech and transcription models through the user's own Vercel AI Gateway access; generate images from text and references, add voiceover, transcribe sound or create video shots. A `direct` shot recipe retains a film budget across attempts, reviews each take before acceptance, and can use an accepted clip's last decoded frame as the next shot's reference. - Footage editing and delivery: import existing footage or recording bundles; remove pauses and filler words, align sound, reframe speakers, zoom into screen actions, add captions, graphics, color and audio treatment; export clean and captioned versions in 16:9, 9:16, 1:1 and 4:5 from the same edit. Recording bundles can carry clicks, cursor movement, key activity and focused-input information. Installation is via Bun 1.3.14 or newer, with a verified v3.2.6 release archive and a `slopcamera skill install --target agents` step; contributors can build from source. The README notes that the source tree prepares v3.2.7 while the install instructions stay on v3.2.6 until publication completes. Native engines install separately, and native Python authoring requires explicit current-user trust because it runs as the current user. The README is explicit about limitations: runtime support varies by platform and profile; interchange preserves only a supported subset, so native rigs and simulations do not become editable Three scenes by exporting a GLB; generated media requires review because models may change subject identity, motion or text; trusted code is not sandboxed; and the MCP server is a subset that checks and renders diagrams, vectorizes images and generates images but does not expose every local CLI operation. On trust and privacy, the README states there is no account, hosted project database or browser generation service. Ordinary editing and rendering remain local. Gateway generation and selected cloud analysis use credentials from the local process and require explicit acknowledgement before uploading named media. Original media remains unchanged by normal edit operations, projects retain explicit revisions, and important operations record inputs and outputs. The README also warns that native tools, providers, codecs and GPU drivers can affect results, so retained source identity alone does not promise identical pixels on another machine. For integrations, the SDK is importable from `@hraness/slopcamera`, with subpaths for declarative workflow graphs, trusted Bun workflows and the local media engine. Verification is done with `bun install --frozen-lockfile --ignore-scripts` and `bun run check`, which the README says covers public SDK boundaries, the local runtime, schemas, Agent Skill, generated entrypoints, static site, deterministic and property tests, and packed consumers. The project is MIT licensed with third-party notices in NOTICE.